python-voiceio 0.9.3__py3-none-any.whl → 0.9.4__py3-none-any.whl

{python_voiceio-0.9.3.dist-info → python_voiceio-0.9.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-voiceio
-Version: 0.9.3
+Version: 0.9.4
 Summary: Voice dictation for Linux. Speak → text, locally, instantly.
 Author: Hugo Montenegro
 License-Expression: MIT

{python_voiceio-0.9.3.dist-info → python_voiceio-0.9.4.dist-info}/RECORD

@@ -1,5 +1,5 @@
-python_voiceio-0.9.3.dist-info/licenses/LICENSE,sha256=Gz61o8jFTAvZUZyB3nWDB3DQQVuipjfPkVu9W8hBHM0,1072
-voiceio/__init__.py,sha256=xKd3pzbczuMsdB08eLAOqZDUd_q1IRxwZ_ccAFL4c4A,22
+python_voiceio-0.9.4.dist-info/licenses/LICENSE,sha256=Gz61o8jFTAvZUZyB3nWDB3DQQVuipjfPkVu9W8hBHM0,1072
+voiceio/__init__.py,sha256=e56AvHfJCtG2ZwwINqsxINVbehWdKxMYgIDbjd7P-II,22
 voiceio/__main__.py,sha256=xT5QCGGreYMisHO7Lh_Y-xAQ2TOzG2D7npKknGNcWSY,53
 voiceio/app.py,sha256=8xS_MySy6cvpxXDcd0-6YwQK6KW_iT8Vdf6FqzYUXpI,74743
 voiceio/audit.py,sha256=p4F2BBX4wEC9eS8MSplzfXL52XC2dyIKbgrRvR-WbTU,15802
@@ -26,7 +26,7 @@ voiceio/platform.py,sha256=f_8MmC5q1sCEuEPyKjRqlVvV9FExuGmwn_aJXhkpiEk,11809
 voiceio/postcorrect.py,sha256=iyO7MPIR4a4XIASEGOs9Y3iUXVTfwtVkhVOWj2vY1lk,11201
 voiceio/postprocess.py,sha256=9grkq3I6XgiQV6ZQbIw2_fOUe57tzC7GBPvLXqhhdTs,2979
 voiceio/prompt.py,sha256=9iza8KoQxvO_U9NPMJ8Ep7j-zkIs9nW8N5sacCRpsEk,2904
-voiceio/recorder.py,sha256=x3WG-3GXvqHb0yagXXVeDgjD5PnG6xv6MyRHMTnB68w,13885
+voiceio/recorder.py,sha256=csqm4Q_GuQyDxhJAVDCL6TQd_nna77sfWNiwqF6TsQk,15003
 voiceio/retention.py,sha256=i_VWPCyykT4kQ1uKilTplpiIrACTYljn09BU1vL71A0,6019
 voiceio/service.py,sha256=1YhnDPHnu0Aw66uwsp05TWfs18bf3io_h6r8eNla7nA,12525
 voiceio/snapshots.py,sha256=bm-lzIeYucDUDlPBZiFiWd-jyB2I6Ft1jRANZCO6go4,2351
@@ -74,8 +74,8 @@ voiceio/typers/pynput_type.py,sha256=DTtkT59M-EKOAlTNkvcZrRvf_Fq52OpI8S5t_8S-QAY
 voiceio/typers/wtype.py,sha256=d1wG-HZdYDQxXIysV_Xvk9HixDpaJgl3VEoKbP5vhIs,1786
 voiceio/typers/xdotool.py,sha256=dh1zhPqUT8ihJahJ7ZKm6PtfYY087UAzADx664DvQOM,1356
 voiceio/typers/ydotool.py,sha256=dt0W9ot__W8LQG2so8vZLGIw9leK2Q6g3i-Voo0LXE4,5327
-python_voiceio-0.9.3.dist-info/METADATA,sha256=h-f14EX3kSMP26A2Y12MJbprM_03l0TkRSm0wIYWrao,15877
-python_voiceio-0.9.3.dist-info/WHEEL,sha256=K260EYznzXsJYBQGqmI8VTxEdiZYNvDZwW9cBh9-_MA,91
-python_voiceio-0.9.3.dist-info/entry_points.txt,sha256=U64fA65zxzyLoC8bgbn2ztQVWHLsc0o0H0qCk1J9DMc,218
-python_voiceio-0.9.3.dist-info/top_level.txt,sha256=piwtn309lD6uexQyXdZ-efAVBJF9y6Wfr48Z-8zkNhg,8
-python_voiceio-0.9.3.dist-info/RECORD,,
+python_voiceio-0.9.4.dist-info/METADATA,sha256=jwrZH_VW6__XC_VKHFMHr052Iadetbx3V5Ma26UYZNg,15877
+python_voiceio-0.9.4.dist-info/WHEEL,sha256=K260EYznzXsJYBQGqmI8VTxEdiZYNvDZwW9cBh9-_MA,91
+python_voiceio-0.9.4.dist-info/entry_points.txt,sha256=U64fA65zxzyLoC8bgbn2ztQVWHLsc0o0H0qCk1J9DMc,218
+python_voiceio-0.9.4.dist-info/top_level.txt,sha256=piwtn309lD6uexQyXdZ-efAVBJF9y6Wfr48Z-8zkNhg,8
+python_voiceio-0.9.4.dist-info/RECORD,,

voiceio/__init__.py

@@ -1 +1 @@
-__version__ = "0.9.3"
+__version__ = "0.9.4"

voiceio/recorder.py

@@ -15,6 +15,11 @@ if TYPE_CHECKING:
 
 log = logging.getLogger(__name__)
 
+# Peak amplitude below this is treated as no signal at all (a muted mic emits
+# exact zeros; a working mic's "silence" still noises above ~1e-4). Matches the
+# floor in has_signal(). Used only to detect a dead mic, never speech vs pause.
+_SIGNAL_FLOOR = 1e-4
+
 
 class RingBuffer:
     """Fixed-size ring buffer for float32 audio samples."""
@@ -111,6 +116,10 @@ class AudioRecorder:
         self._no_speech_secs = cfg.auto_stop_no_speech_secs
         self._sustained_silence = 0.0
         self._heard_speech = False
+        # Raw-signal (amplitude) tracking for the muted-mic auto-stop, kept
+        # separate from the VAD-based _heard_speech / _sustained_silence above.
+        self._heard_signal = False
+        self._silent_signal_secs = 0.0
         # Called with a reason: "silence" (spoke then paused) or "no_speech"
         # (nothing ever heard — a muted mic or a forgotten hotkey).
         self._on_auto_stop: Callable[[str], None] | None = None
@@ -221,6 +230,8 @@ class AudioRecorder:
             self._silent_chunks = 0.0
             self._sustained_silence = 0.0
             self._heard_speech = False
+            self._heard_signal = False
+            self._silent_signal_secs = 0.0
             self._pause_fired_at = 0
             self._meter_peak = 0.0
             self._meter_clipped = 0
@@ -301,8 +312,19 @@ class AudioRecorder:
 
         # Level metering
         abs_chunk = np.abs(chunk.ravel())
-        self._meter_peak = max(self._meter_peak, float(abs_chunk.max(initial=0.0)))
+        chunk_peak = float(abs_chunk.max(initial=0.0))
+        self._meter_peak = max(self._meter_peak, chunk_peak)
         self._meter_samples += len(abs_chunk)
+
+        # Raw-signal presence, independent of the VAD. A muted mic delivers
+        # digital zeros; real audio (even speech the VAD fails to classify)
+        # sits well above the floor. This — NOT the VAD — gates the muted-mic
+        # auto-stop, so a recording is never cut while real audio is coming in.
+        if chunk_peak > _SIGNAL_FLOOR:
+            self._heard_signal = True
+            self._silent_signal_secs = 0.0
+        else:
+            self._silent_signal_secs += frames / self.sample_rate
         pinned = abs_chunk >= 0.99
         if np.count_nonzero(pinned) >= 4:
             runs = np.convolve(pinned.astype(np.int8), np.ones(4, dtype=np.int8), "valid")
@@ -345,18 +367,18 @@ class AudioRecorder:
             log.info("Auto-stopping after %.0fs of silence", self._auto_stop_secs)
             cb("silence")
 
-        # Safety net: nothing heard at all for a long time. Because speech
-        # resets _sustained_silence AND is gated out by not-_heard_speech, this
-        # measures continuous silence from the very start of the recording — a
-        # muted mic (all zeros) or a hotkey pressed by accident. Distinct reason
-        # so the app can surface it instead of failing silently.
+        # Safety net for a dead mic: NO raw audio signal at all for a long
+        # time — the mic is muted / unplugged / delivering zeros, never merely
+        # that the VAD didn't flag speech (that would cut real dictation off
+        # mid-sentence). _heard_signal latches on the first non-zero chunk, so
+        # this only fires for a mic that was silent from the very start.
         elif (self._on_auto_stop is not None
                 and self._no_speech_secs > 0
-                and not self._heard_speech
-                and self._sustained_silence >= self._no_speech_secs):
+                and not self._heard_signal
+                and self._silent_signal_secs >= self._no_speech_secs):
             cb = self._on_auto_stop
             self._on_auto_stop = None
-            self._sustained_silence = 0.0
-            log.info("Auto-stopping: no speech heard in %.0fs (mic muted?)",
+            self._silent_signal_secs = 0.0
+            log.info("Auto-stopping: no mic signal in %.0fs (muted?)",
                      self._no_speech_secs)
             cb("no_speech")